VRay and others on GPU - Siggraph 2009

[owei]

Chaos Group V-Ray RT GPU Rendering Demo - SIGGRAPH 2009 Recorded by CGarchitect.com

Take a look at the video here:

http://www.cgarchitect.com/news/SIGGRAP ... -GPU.shtml

..and the news on the Chaossoftware homepage:

http://www.chaosgroup.com/en/2/news.htm ... #newsid126

and an other video of the presentaion for download:

http://www.spot3d.com/vrayrt/gpu20090725.mov

cheers,
oliver

[calmasacow]

Yeah This totally bummed me out. How is it that Nvidia has Many time more resources than ChaosGroup and yet still no Hint as to when we will see a GPU server of Mental ray which they own. I think they fear loss in the sale of standalone licenses. I think ChaosGroup said that it would be at least another year before this tech was ready for primetime. I think that Nvidia will wait as long as possible to release a GPU version of MR. if people can render 7-6 frames per second on a GPU it is less likely that they are going to buy hundreds if not thousand of these standalone licenses that they are buying now.

Of course Nvidia has the Bigger advantage that Mental Ray is already in just about every package and they have the ability to sell MR-Cuda Standalone bundled with a Tesla GPU card.

But unless Nvidia acts soon this will mean bad things for Softimage and Maya as well. ChaosGroup has no Vray RT "what Vray GPU is based on" even in development for any other platform. With the cost saving that can be gained from this I fear that many studios will feel the pressure to move to Max.

[owei]

I agree particially. Small firms seems to work much more effective and much more innovative than the big players.
As far as GPU-raytracing is concerned, NVIDIA has shown their GPU renderer at Siggraph:
http://www.nvidia.com/object/io_1249366628071.html
This one is a unbiased renderer and it´s not called MR or something. And, it only works with quadros. That means of course NVIDIA wants to sell hard- and software.
Never the less. In the chaosgroup presentation no textures were shown. This might become a bottleneck, meaning that the transfer from cpu-ram to gpu-ram takes time and has to be managed. Looking forward to chaosgroup´s presentation at the "softimage uebertage" usermeeting next month. Perhaps, there will be some news about the softimage version of VRay.

cheers,
oliver

[calmasacow]

Still, I think think that a whole new engine is not the route to go especially when you own the code base to the most widely used rendering engine on the planet. It make more sense to port MentalRay to run on the GPU than it does to create another whole platform that everyone will have to adapt their workflows and code bases to. I think it is a bad business decision. Not saying that a new approach is not needed. "mental Ray is showing it's age." But I think that accelerating MentalRay first would have been a better business strategy. Then worry about the next generation render.

[owei]

Have a look at a new hardware raytracing accelerator card called "Caustic one".
They will try to get industry standard for gaming and dcc..we will see ;)

http://caustic.com/caustic-rt_intro.php

cheers,
oliver

[calmasacow]

Have a look at a new hardware raytracing accelerator card called "Caustic one".
They will try to get industry standard for gaming and dcc..we will see ;)

http://caustic.com/caustic-rt_intro.php

cheers,
oliver

yeah not really encouraging seeing how the only application I could find even mentioned was 3ds max.

Again seeing how MEtal ray is in so many apps it makes more sense to make a Cuda accelerated version of the Mental Ray Binary that can be dropped in as a direct replacement for any application using that version of mental ray.

[origin]

caustic one seems to be a better option, it produces final-quality images

we have vray RT at our work and its only a preview tool.

Andreas posted also a little video of Arnold interactive rendering which seems to be a mental-vray-brazil-finalrender killer

[calmasacow]

ha yes the often spoke of but never released Arnold renderer. I think they have gone on record with saying that they are not looking to sell to anything less than a studio buying 30-40 seats at like 5K a pop. I wish them luck but I dont see them ever getting that. Renderman doesn't even demand that anymore. Speaking of which that is something that would be really sweet. GPU accelerated renderman. I would love to have that running on a quad Tesla machine.